Inetd svc.configd hogging cpu

Question

I'm running a tcp proxy for pop/imap on a v210 with solaris 10 and zones.
The tcp proxy accepts connections on ports 110 and 143, does some datbase lookups on other servers and then forwards the connection to the appropriate servers.
The tcp proxy is running under inetd inetconv'd to smf. Everything runs fine until the connection rate reaches aproximately 25 connections/second.
When this rate is reached cpu is 100% busy (about 50% user and 50% sys). Dtrace shows that svc.configd is very busy doing llseek calls.
I tried to increase stack and heap page sizes (in /etc/default/init) to 4M. This resulted in cpu loads of 30%user 70 %sys, but still 100% total.
The tcp proxy processes are ver short-llived so it seems configd is hogging the cpu to keep up with the state of the procsses.
Is there any way to tell configd not to keep up with these processes, or alternatively to tune the system to increase configd's performance?
I want to reach a connection rate of about 60-80 connections/second.
Leo.

Answer 1

I'm trying a workaround now:
adding
<property_group name='startd' type='framework'>
<propval name='duration' type='astring' value='transient' />
</property_group>
to inetd's and the service's manifest seems to improve performance. The cpu load doesn't get up to 100% anymore, though still high (about 50% total). Awkwardly the setting doesn't show up in svcprop, but behaviour is definitely better.
Leo.