One more...

>I've tested operation of InProc and OutOfProc servers with Web
>Connection using its pool thread manager and the difference between
>InProc and OutOfProc is marginal - the main reason is that each request
>is a single method call to the server, but even with 15-20,000 hits an hour testing under load there was only a 2-5% difference.

If you now throw MTS into the mix you loose about 10% of performance from
the EXE...