Investigate deserialization performance annomalies

[merdacz]

Please refer to LoadTests unit-test for additional information on the individual scenarios tested.

Inverted processing times

While in most cases alternative version is superior to original one in that particular one it persistently is not. The guess is that it may have to do with GC collection. Forcing GC may help overcome this bias.

Version	# of entries	# of loops	Total time [ms]
Original	10	500	505 *
Alternative	10	500	897 *

Similar total size yields different total times

While this is not as unexpected as the above, it may still be worth looking into.
While total "effort" (entries times loops) is the same, lower number of loops (external for) yields better
result compared to higher count one with smaller number of entries. It may have to do something with buffer allocation/GC collection as well.

Version	# of entries	# of loops	Total time [ms]
Original	2	5000	1548
Alternative	2	5000	1192
Original	20	500	919
Alternative	20	500	718