Presumably something like this, but probably more sophisticated:
Bing sidebar extracts list of unique words appearing on each site user goes to and feeds back to MS.
MS correlates keywords to the next site that the user visits.
Google honeytrap employees search for trap term on google, click on the returned link which goes to an unrelated site.
Since the honeytrap term is unique and no site contains this word, the only data the MS has that relates to this word comes from the sidebar data stream which shows that when a user visits a page containing this word, the next page they go to is the unrelated site. Thus the make the connection that that page is relevant.
I would have thought however that MS would have put a check in to not pull data from the sidebar when the user is visiting Google to precisely avoid this kind of accusation. Since Google found the issue for some of the words in the honeytrap but not all of the words suggests that something more subtle is happening. For example, some of the honeytrap employees might have checked their email using an online hosting service (e.g. gmail etc) to find out what the honeytrap word is before proceeding. The sidebar extracts the unique words from the email and sends to MS. Two or three clicks later the employee is at the unrelated site. Suppose the sidebar is setup to ignoring Google searches. If MS correlates to a few links out it will see the honeytrap word from the email related to the unrelated site without observing the Google results.