There are definitely some drawbacks to MVCC. For example, many people are surprised how long "select count(*) from X" takes to run, because it has to do a full table-scan to check which rows the current transaction can actually see. You need to periodically 'vaccuum' tables, which updates a cached list of which table rows are deleted, so they can be reused by future update (modern Postgres will auto-vacuum, but it's another set of knobs for the DBA to tweak). Plus, the storage overhead mentioned by the article (although that really depends on the amount of churn in your particular application, and regular vacuuming keeps it to a minimum).
It is kinda funny that you used the phrase "pick your poison". That's exactly my feeling when working with SQL Server. I have never had a "pick your poison" moment with PostgreSQL :)