Generally speaking I'd go with TBB. Of the three it's the only one that does both single and multiple thread tests. BTw I would ignore. My results aren't as bad as yours but it's never been very accurate.
However I'll also point out that the you are testing the transfer speed between you and the remote server. Unless the servers are located in the same location (which they aren't) your data is going to be travelling across different parts of the network. It is entirely reasonable for their results to differ. In the same way that if you measured your average speed driving to Exeter it would be unusual for it to be the same as your average speed to Edinburgh. Those averages could change at various times and sometimes it's higher for Exeter, sometimes for Edinburgh.
It's best I think to pick a collection of testers and run them all regularly. That way you're looking for changes in normal behaviour rather than just looking at an almost arbitrary headline speed.