step2: port a beefier version of the architecture + chipset to another socket around 6 months later
step3: die shrink step1 around 12 months later
step4: die shrink step2 around 18 months later
repeat from step 1
and these are the rules:
multi thread performance: step4 > step2 > step3 > step1
single thread performance step4 = step3 > step2 = step1
they have been doing that for the last 6 years and by all indication will keep doing so.