The TLAB rule was very primitive. It simply looked at the ratio between the stuff allocated inside of TLABs and total allocated. This is only interesting if there is a high allocation pressure. So, a new rule should take into account:

1. The allocation ratio (Inside TLAB / (Inside + Outside TLAB)).
2. The allocation pressure (time spent in GC / total runtime perhaps)
3. The likelihood this is something that can be addressed at all by changing TLAB size. I.e. if there are a few huge allocations outside of TLAB, that will not be affected much by changes to the TLAB size. OTOH, many allocations outside of TLABs, with a size reasonably larger than the current TLAB size is an easy win.