Using tapes on tuplesort.c

Hi to all.

It seems that the current PostgreSQL implementation of the Replacement Selection (RS) algorithm [Knuth] changes a logical tape for each run built.
I'm trying to implement that refinement to RS using 2 heaps instead of just one (2Way RS). Recall each heap is aimed at building its corresponding physical run, both heap cooperate building its own physical run associated to the same logical run).

"2Way RS" stops building the current logical run just after stop building both physical runs associated to the current logical run.

My question: should I use/change tape for each physical run or for each logical run?

I know you'll be probably busy with issues on the new PostgreSQL release, so I'll thank you twice for your reply.

Regards, Manolo.