Whilst performing administration tasks on a clients BizTalk 2006 environment the other day I noticed approximately 25,000 service instances in the active state. Pretty hard to miss especially with the size of the spool table. The service classes were a number of send ports from the same application all configured to run under the same host instance. That host simply locked up and stopped processing. First port of call were the BizTalk performance counters. Specifically the BizTalk:MessageAgent Message delivery throttling state counter which was reporting a throttling state of 3.
The send ports had received over 25,000 service instances in the matter of a few seconds which resulted in the host configuration setting In-process messages per CPU threshold being breached tripping the throttling state. The host was running with the default out of the box settings. A bit of analysis on what the value should be set to followed by a bit of tuning and a host instance restart saw those 25,000 mesasges processed quickly. More information on throttling states can be found here, http://msdn.microsoft.com/en-us/library/aa559893(BTS.10).aspx