i am not sure what merge tool did, but theoritically or if you see the code
it will use memory of 2x staggersize x 64 byte, so 1M stagger will use 128 MB of ur memory for each plot
on the console did the miner said "done reading plots .... xxxxx nonces", how many nonces did it read, does it equal to the number of nonce your plot file has?
or if you see difference between miner calculated best deadline and confirmed deadline (pool calculated deadline) it probably your plot is damaged, has funky scoop alignment, or your plot data does not same as your filename describe (accountid, nonce start, number of nonce or staggersize)
it read all of them, i think the version of my stagger optimizer isn't working right, i'm re-doing the plot with the new cpu plotter, which is much faster and can create better stagger without the need to optimize them