As Dr. Wheeler explained very nicely in his books, sigma is a global measure of dispersion which does not take into account the time behaviour of the data and its "between subgroup" variation, even if we are talking about subgroups of 1 in individual charts.
2.66 * MR consider the data MR and therefore is dependent upon data time distribution.
This is the reason why, if you use limits caluclated with 3sigma, you run the risk to catch the wrong signals (or miss the right ones) coming for your process.
The attachment may help you to clarify this subject.
Ciao!
Giuseppe