Hi Kevin,
Doubling is a bad idea. You can do this in PALM: Make the design as usual, but include subject-specific intercepts for all those that have two measurements, that is, include one EV per subject that is all-zeroes, except that it contains +1 for one run and -1 for the other run.
Then define multi-level exchangeability blocks: one block per subject in one level. In the level above, one block for all subjects with one run (so these subjects can be permuted with each other), and one block for all subjects with two runs (likewise, these subjects can be permuted with each other). Finally, a top-level that prevents shufflings (with a negative sign in front of it).
There is no need for variance groups.
Hope this helps!
All the best,
Anderson