weird Yin~ behaviour

I get some weird behaviour when I use the downsampling feature of yin~. I have three subpatchers that have a yin~ with different downsampling arguments (yin doesn't seem to take any messages). Then I use pcontrol to enable or disable the subpatch.

Anyway, the weird thing is that if I use downsampling 2, my CPU is almost twice compared to downsampling 1. Downsampling 3 is a bit less than 1. The even weirder thing is that once in a while I will get more "expected" results of cutting CPU in about half for each time I downsample... That seems to happen the first time I use yin~ after Max startup.

Can anyone shed some light on this?? Thanks

Below is an example patch. I have yin~ version 05/2007 (6).