As mentioned in some postings above I've now changed the motherboard as well as the memory, but the result stays the same.

I think, I'll cancel my subscription to FAAH but will crunch for the other WCG projects because they work fine on my PC.

I'm having trouble with invalid results as well.
Machine ID 89494
faah1043_ d252n140_ x1AJX_ 02 squid Invalid 12/06/2006 11:30:03 12/07/2006 08:14:31 6.12 69 / 41
faah1043_ d252n659_ x1AJX_ 00 squid Invalid 12/06/2006 05:26:20 12/06/2006 14:47:03 2.66 30 / 30
faah1042_ d251n688_ x1AJX_ 00 squid Invalid 12/05/2006 17:48:22 12/06/2006 11:30:02 5.24 59 / 36
faah1042_ d251n012_ x1AJX_ 00 squid Invalid 12/05/2006 17:43:10 12/06/2006 05:14:16 7.04 74 / 42

There's more marked as Inconclusive that I'm pretty sure will be Invalid. I haven't managed to get a single Valid result, but there's more to the story...
Every one of the Invalid results has warnings like this in the log:

call_glss(): pop_size: 200 num_evals: 10000000 start: [18:27:41] 
WARNING! The population appears to have converged, so this run will shortly terminate.
call_glss(): end: [18:27:42] 
wcg_checkpoint() called
Skipping checkpoint
call_glss(): pop_size: 200 num_evals: 10000000 start: [18:27:42] 
WARNING! The population appears to have converged, so this run will shortly terminate.
call_glss(): end: [18:27:42] 
wcg_checkpoint() called
Skipping checkpoint

An installation on another machine which has no problems returning Valid results never has these messages. The problem machine is not overclocked, I have even tried underclocking it with no success. Heat isn't an issue, I do a lot of video encoding with xvid, x264 etc. which raise the temp higher than BOINC and there are no problems there. I ran memtest overnight and cpu tester the following night with no errors reported.
There are no virus checkers and external firewalls installed. Both machines access the internet through the same router.
I've run a dependency walker on wcg_faah_autodock_5.28_windows_intelx86 on both machines and compared the dlls loaded and all have the same versions, so it's not a windows update issue.
The problem machine is set up to dual boot between regular windows and windows x64. If I run the client (the same installation) under windows x64 there are no problems (the convergence warnings do not appear in stderr.txt). Further proof it's not a hardware issue.
I've tried resetting the project, no difference.
FA@H is the only project that has issues, HDC and GC are fine.

I'm pretty much out of ideas. Anyone else got any?

It certainly appears that the WUs processed under x64 are ok. I will confirm when they're validated.
All the Invalid units have fourth copies, my results are Invalid while the other 3 are valid. My times are slightly shorter which I'm guessing is because dockings are being falsely reported as converging.
I only booted into x64 to check, I don't run it regularly. The client had been running in x32, I copied stderr.txt from the current WU to a backup location, deleted the checkpoints and maps so it would start from 0% and started boincmgr.exe. I let the client run for a while then checked the new stderr.txt against the old one from when it was running under x32. The old one had lots of convergence warnings, the new one didn't.

I might have found the cause - so far it's done 7 dockings without a warning which is further than it normally gets. I'll keep you posted.

Thanks for letting us know.... MSN is supposed to use ports 1863, 5190, 6891-6901 (6667+28000 for games) per the www.portforwarding.com site. If u go there, u can find guided help how to configure e.g. your specific firewall and/or router. Here an example of how to set up MSN with Kerio PF... http://www.portforward.com/english/routers/fi...ewall/MSN_MessengerIn.htm

Been using Trillian for years, German manufacture, Grundlich, multi protocol AIM, ICQ, IRC, MSN, Yahoo... not the latest hi-tech screamer, but never a twist either. There's a free and a paid version.