My understanding is the there are two stages to get to a win.
Stage 1 gives you a 1 in 255 (or something like that) chance of a plot passing the filter. So the more plots you have, then the more plots pass the filter. Only plots that pass the filter move on to stage 2.
Stage 2 runs some type of proof. Whoever has the best proof wins the XCH.
So far, it seems like K32 would give you the best bang for the buck, because way more K32 plots could qualify for stage 1, given a the same finite amount of disk storage for all of your plots.
But each plot, as best I understand Chia, contains multiple sets of data within. So whichever part within a plot has the best proof will win.
This means that larger plots (K33 and K34) have 2x or 4x the chance of finding a better proof within their file.
So although roughly ½ the amount of K33 plots will pass stage 1, compared to K32 plots that will pass stage 1…
…the K33 plots that do pass the stage 1 filter will have 2x the chance to win when it goes up against a K32 plot that passed the stage 1 filter.
So it all kind of equals out. With K33 plots and K34 plots, you get ½ or ¼ the number of plots that pass the stage 1 filter. But the ones that do pass have 2x or 4x the chance of winning vs the K32 plots that pass the filter.
This begs the question as to why bother with the larger plots when they do not really offer better odds of winning?
From where did you get 10 years?
When the day comes that K32 plots get phased out and will one day no longer be supported by Chia, then it will be a mad rush to re-plot.
If you have hundreds of TBs of plots, you will have possibly months of re-plotting to do. I am doing it now.