ORA-600 [KCBLASM_1],  on Patchset 10.2.0.5
During routine alert.log check control I spotted several ORA-600 errors:
ORA-00600: internal error code, arguments: [kcblasm_1], 
It is a known issue introduced with 10.2.0.5. This platform generic bug (7612454) is a regression bug introduced in Patchset 10.2.0.5 according to MOS note:
Bug 7612454 – More “direct path read” operations / OERI:kcblasm_1 [ID 7612454.8]
Obviously, bug 7612454 causes performance problem with direct path reads, compared to performance from before (10.2.0.5). [Right now, I can’t confirm how severe performance impact is, because no one so far reported a performance problem or any crashes.]
Since no workaround was mentioned in above MOS note I searched further and found related note that highlights the problem a bit more:
ORA-600 [KCBLASM_1] RUNNING A QUERY WITH HASH GROUP BY [ID 848094.1]
At the time of this writing I’ll probably disable group by hash aggregation at instance level (_gby_hash_aggregation_enabled = false), if the problem becomes severe. So far, this bug is more of an annoyance than a show stopper – but I feel better being prepared with a workaround. I hope we’ll not hit the case when neither setting _gby_hash_aggregation_enabled = false nor _hash_join_enabled=false helped as a workaround, as described in MOS note Bug 9918715: ORA-00600 [KCBLASM_1] ERROR.
The second suggested workaround, upgrade to 11.2, is (of course) out of the question at this time. Besides, who can guarantee us that we will not hit some XY regression bug in 11.2 the day after we go in production with 11.2! ;-)
What to say for the end, 10gR2 deserves to stay written in my black book with golden letters as being the buggiest version ever released by Oracle. Someone can argue that bugs goes hand in hand with features, software business as usual; I had to say, with every new grey hair I have less and less tolerance to excuses like this.