Hi Dilip
At my work place we poll MDA table monProcessActivity to zero down to SPIDs with high number locks using column called LocksHeld. In fact we do not allow more than 1M+ locks for a SPID. SPID with more than 1M+ is terminated during the poll and DBAs are notified with SPID + KPID + Login details.
We have third party tools collecting SQL so we use those to get actual SQL/Plan if need be.
But the same can also be extracted from monProcessSQLText or good old dbcc sqltext(spid) before the SPID is sent to a better place !!
What I have observed is that number of locks climbs extremely rapidly when user(s) use "select into" for creating temporary tables while extracting data from a very active table with datarows locking scheme and also with huge number of rows. Most probably server can not lock the table(s) due to high activity and the number of locks keep increasing to cross the configured limit.
I would suggest not to increase the lock limit but get end users to agree on some reasonable number of locks and then use that as hard limit to terminate a SPID. If you provide enough evidence to end users they normally agree because no one wants to see a worst case scenario of recycling a server during a busy day.
In summary MDA tables are your friends. Use them positively to help you.
HTH
Avinash