Describe the bug

Upto 20% SET command regression due to IO-Threads redesign cleanup work #3544

Check HERE
The regression can be seen for IO-thread test cases for SET commands.
This does not cause a regression for 16 byte value, but does for 128 and above

I feel this would be the root cause for the regression.

• tryOffloadFreeObjToIOThreads: only offload the free of the sds buffer
  (which the I/O thread allocated). The robj itself is main-thread-owned
  and is now freed on the main thread.