The Google paper you're referring to is here[1]. Their results show that very high temperatures do affect older drives (> 40ºC, 3+ years), but temperature variation within lower and mid ranges shows little correlation with failure rates.
The important result from Google's and others'[2] research is that lifespan for particular drive models is essentially bimodal, with the first peak during the first couple of months - strongly correlated with manufacturing batch, not just make/model - and the second peak after 3+ years due to age related causes. It's possible that people who experience entire arrays of drives failing within the first few months of use have chosen a particular batch of drives that suffer from high 'infant mortality'. One recommendation is to build arrays from drives with different manufacturing batch number or - if the RAID implementation allows it - use completely different makes and models.
The important result from Google's and others'[2] research is that lifespan for particular drive models is essentially bimodal, with the first peak during the first couple of months - strongly correlated with manufacturing batch, not just make/model - and the second peak after 3+ years due to age related causes. It's possible that people who experience entire arrays of drives failing within the first few months of use have chosen a particular batch of drives that suffer from high 'infant mortality'. One recommendation is to build arrays from drives with different manufacturing batch number or - if the RAID implementation allows it - use completely different makes and models.
[1] http://research.google.com/archive/disk_failures.pdf [2] http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=1285...