In the 10.2.0.1 version of the oracle database a new heap checking mechanism, together with a new messaging system is introduced. This new mechanism reports memory allocations above a threshold in the alert.log, together with a tracefile in the udump directory.
In certain situations it can be handy to know if large allocations are being done in the sga heap (shared pool), but only to troubleshoot memory allocation problems (which eventually will appear as the infamous ORA-4031). For normal day to day work, I doubt the effectivity of these messages.
The default limit is set at 2048K. It appears normal usage can result in messages from the memory manager. During my last installation (with all options set default, all options installed), this resulted in 125 messages from the memory manager.
Oracle gives a solution on metalink in note 330239.1:
Set _kgl_large_heap_warning_threshold to a “reasonable high” value. This parameter is set in bytes, and oracle recommends the value 8388608 (8192K).