Final Cut Express occasionally renders audio "out of order"

Question

I'm running into a very occasional problem, which is just rare and random enough to not be recreatable on demand, and so is proving difficult to troubleshoot.
In a nutshell, sometimes - not very often - when I render an audio stream it will report an estimate of 6 hours or more for something even as small as a 1-second transition.  This is not a composite or layered video or audio by any means, just a standard cross-disolve/cross-fade (+3) between two adjacent clips.  When this type of extreme estimate occurs, I may find afterward that rendered audio in the sequence has been moved around, sort of like a shell game.  Usually this anomaly only happens over a minute or less of the project.  Trying to describe this, a portion of the rendered audio from one part of the timeline will replace a portion of rendered audio elsewhere -- as if some audio render segments have gotten linked out-of-order.  If memory serves, the displaced audio so far has always appeared earlier in the timeline than it should.  The replaced audio just doesn't seem to appear anywhere.  So far, if I remove the associated audio render files and re-render, it fixes the problem, but I'm trying to sort out what is causing the behaviour in the first place so that I can take steps to keep it from happening at all.  It raises havoc with trying to meet deadlines when I'm stuck re-rendering sometimes entire segments....
For some details about the environment:
     System is a Mac Book Pro, 2.4GHz Intel core duo, 4GB RAM, OSX 10.5.8.
     I'm using FCE4, importing SD clips for TV production from Sony cameras (sorry, I'm not part of the camera crew, and I can't recall the exact model number of their cameras - they are all the same model though, configured the same, so no known variables from that end).
     The cameras record standard-def MPG (Mpeg layer 2, so yes I convert them before importing into FCE), I convert them using the ffmpeg library.  I tried FFMPEGX, but I find I have far more control over video/audio parameters and combinations with the command-line version.  The parameters I use result in a Quicktime (.mov) file with these parameters: 16-bit integer (little-endian), stereo, 48kHz, DV 720x480 (640x480), Millions (of colors).  FCE is generally quite happy with these parameters, and does not require immediate rendering of the audio or video - unless plugins are applied of course.  As far as plugs, I generally only use the built-in FCE plugs for comp/lim and parametric EQ.  If I have to get much more detailed than that, I'll go to Digital Performer (read on).
     In some cases the audio of the original clips is in quite a bad state, and requires far more detailed restoration than FCE can provide.  As I am primarily an audio engineer, I rely on Digital Performer and friends for audio restoration, exporting from there a new clip, usually duplicating the original video (self-contained), though sometimes I do let DP export a reference movie.  Final Cut doesn't seem to have any trouble with the .mov files exported from DP either way, and this anomaly doesn't seem to care whether the clips involved were processed by DP or not.
     Disk space, memory, and CPU load are all well within acceptable ranges; and with my previous years of experience as a systems admin, I am well aware of the system impact and ramifications of using internal vs external disks, memory allocation, disk filesystem operating overhead, etc.
As I've mentioned, this is not a show-stopper, I have a work-around to correct it when it happens.  It's just a bit of a bugger *when* it happens, and I really wish I could work out the root cause of it, so that I can take steps to avoid the problem in the first place.  Any suggestions/insight/help would be greatly appreciated, thank you!  :-)

Answer 1

Thank you for your response.
Grr.  I had a nice response ready to post, and apparently the system logged me out and wouldn't let me log in while keeping the text.  I'll condense it a bit, since I've stumbled across some additional points to ponder while typing.
The clip:
And the sequence:
Yes, I see that the clip is 16-bit integer and the sequence supports 32-bit float, but this should not pose a problem since modern digital audio processing up-converts smaller sample formats to larger ones on the fly for processing, and exports whatever format is desired at the end.  If this *was* the problem, I should always be off by a ratio of 2:1 on all audio data, but I'm not.
Additional notes: the system generally has Firewire, USB, and network mounts (all required for my work), and the internal drive (which *is* 7200RPM) is partitioned into a system and a user area to keep audio and video projects out of the system's disk space.  Most main clips for a project are placed on the internal user space for speed, externals hold auxilliary files (stills, adverts, livetype, etc).
It was at this point in my previous typing that I began to realise that the anomaly seems to begin in sync with some random livetype object.  It then extends a bit past the livetype object, so it's not locked to that object; and multiple instances of the same livetype object DO NOT result in multiple instances of the audio anomaly; it only hits a single instance; and again it usually cleans up when the audio render files are removed and re-processed.
This leads me to wonder about a possible odd combination of events/processes on the system, and whether or not FCE handles livetype natively or requires a livetype helper (i.e., another process stuffed into the system table).  This sort of external influence would be a transient sort of event, and therefore could easily be out of the way by the time second render attempt is begun - resulting in the "fix" by removing and re-generating audio render files.
It makes sense to me, but it's still hypothesis and supposition at this point.