As Underfire and Asquared point out there are a number of errors or rather inaccuracies that lead to "drift"

You'd imagine that some of them would/should cancel each other out but it never seems to work that way - they always tend to be additative....................

:ok:
Thanks for this

That's because you never know in which direction the errors are. And the number you are after is the worst case, where the errors all conspire together to do you in.

In the pre-GPS 767 I flew the three IRS positions were essentially triple-mixed in the FMS, similar to the triple-mixing we did with three INSes.

The modern stuff keeps the IRS accurate by continuing tweaking the position with the GPS inputs. If all airframes that are RNP AR qualified had these types of IRU/IRS, the containment assumptions for RNP missed approaches could be significantly less than those used in today's criteria. The RNP missed approach assumes loss of GPS and IRU/IRS drift rates of the older units.

Pepperseed, INS DME updating was available provided you had a valid signal from the DME.
EEngr That was the reason for the carouseling, to randomise errors by turning the INU platform at, as I recall it, 4 RPM. It was pretty good with residual errors rarely anywhere even close to the 3+3T limit.

No so with the INS sets in our L1011/B747 fleet.