Only downsides would be cost-per-run pricing and low-latency queries.
One point that this article doesn't address (not that it necessarily SHOULD) is the fact that BigQuery achieves these results with zero database administration or schema design.
Redshift is great, but it's not "set and forget".
Also, there are a huge number of use cases out there which are daily reports on large datasets. Fixed vs. Variable cost models are obviously going to shift in favor of variable when this kind of scenario comes into play.
Last, what kind of cost factor do you think comes into account when it comes to paying someone to administer your Redshift db, vs not paying anyone to administer the BigQuery setup. I can tell you from experience that this is a huge, huge factor.