StringBuffer delete (int start, int end) is too slow - HELP!!

Question

I've traced a performance problem to one line of code: a StringBuffer "delete (int start, int end)" call.
My StringBuffer contains a lot of data, maybe 500,000 characters. My code loops, searching for a particular character (the search is fast, btw). When that particular character is found, I want to delete that character plus the characters following it. My code looks something like:
int pos;
while ((pos = myStringBuffer.toString().indexOf('~')) != -1)
     myStringBuffer.delete(pos, pos + 3);I'm 100% sure the bottleneck is the delete call. If I change the code to the following, where I've commented out the delete, it loops the same number of times and runs lightning-fast:
int pos = 0;
while ((pos = myStringBuffer.toString().indexOf('~', pos)) != -1)
     pos++;The loop, and therefore the delete method, will be executed thousands of times. To give you some idea of the poor performance I'm seeing, the first code (with the delete) takes more than 10 minutes to run, while the second code (without the delete) takes less than one-tenth of one second.
Any suggestions? Thank you.

Answer 1

Thanks for all the help! da.futt's solution works perfectly: 65,000 times through the loop in about one-tenth of one second. Problem solved.
We're still on Java 1.3.1 (large company = slow to upgrade), so I can't even remove toString(). I wonder how many "small" performance problems we have (and are unaware of) that could be addressed by upgrading?
My first posting here and a positive result. I'm not a Java expert, but neither am I a rookie. I'll be back to try and help others as I've been helped here.
Thanks again.