Interesting discussion, I'd like to add my thoughts.
Like many on here my both sides matches jumped from just my sister to about 40, all of which significantly have 2 segments that match.
It seems to me that the phasing falls into two stages, first separate out the base pairs into phased segments, then allocate those segments to a parent. The first part should be relatively easy given sufficient matches, the second part much harder as it would depend on matches having two or more segments in common. If this is correct then the errors are likely to come in at the parental allocation stage.
So then what I see as the underlying data that Ancestry have calculated would be a bunch of phased segments allocated to one parent or other, say for arguments sake about two hundred. Of these two hundred segments 90% are correctly assigned to a parent and 10% (about 20 segments) wrongly (based on Ancestry's numbers).
So then to get a spurious both sides allocation it is only neccesary to match within one of the segments in the correct group and one of the segments in the wrong group. Hence my both sides matches all have 2 segments in common.
Quite why the number of both sides matches has recently jumped I have no idea, presumably changes in the algorithm.
An interesting side effect of this process would be the ability of Ancestry to eliminate a large amount of IBC matches since they would match neither set of parental phased data.