It may be a side effect of fixing the peek4() CPU cycle exploit, but now all “adjusted” functions seem to cost the same as if they were standard Lua API calls.

The following program used to run at 21.6% CPU on 0.1.11f. Now on 0.1.12b it runs at 64.5% CPU!

function _draw()
  cls()
  local i = 0
  for j = 0,10000 do i = bxor(j, shl(i, 3)) end
  print(tostr(stat(1)*100).."% cpu", 2, 2)
end

Fixed for 0.1.12c, thanks

This was a bug in stat(1)'s reporting rather than a change to the 'actual' (pico-8) cpu usage. So carts under 0.1.12b aren't running any slower, but they say they are. And yes -- it's basically showing what the underlying Lua vm costs are.