One trick is to manually add known AcoustIDs to the top of the recording page. I assume you own a few of these recordings yourself? So you can extract those and use them as a reference. Even better if you have examples of both 1947 and 1942 recordings.
Then if you have a “known” 1947 acoustID, you can safely untick that AcoustID from the 1942 recording. And vice-versa.
Often you will find if you add a few, others will follow.
I know little of Bing, so can’t assist in any detail. In this example you have had others walk ahead of you creating the initial split. And they will have learnt much on those earlier trails.
This task is made trickier as both recordings are the same length. But you can also look for other patterns. For example, the label who is releasing the track on a compilation. You will find they will generally stick to using the same version for many years. There will be a pattern as to who uses the 1942 recording, and who uses the 1947 version.
Also watch out for the badly credited cover versions… a messy situation I’ve been learning of in the past few months. Again it is the Label that helps make those stand out.
Another trick, is to get some better feel of Bing’s output by working on the lesser known tracks. That way you’ll soon work out which are the labels to trust. Once you have a clearer picture from that angle, you’ll have more knowledge for the big tracks like this one. This will certainly assist in that search for the original recordings.