Thanks Jeff. You are right about all the points you cited especially point 4. The allocation buffers are indeed imbalanced. We are, in any case, removing tempdb from the cache. It was setup for a different purpose long time back and no longer needed.
We are going to look into 'statistics=hashing' clause. It seems to be very helpful indeed.
However, I was wondering why there was a stack trace. Wouldn;t the error message suffice?
Wondering because, I want to be sure that it is not causing any other issue. Due to all this, the performance has hit and hence the concerns.