The first thing I noticed was that you're viewing the footage at 165%... at that magnification it's going to look bad.
Second thing that comes to mind is the downconvert via your camera: does the field dominance match the SD footage? If not, capture it in its HD format and convert in Compressor.
You call to take everything to SD was wise, and if it matches in all details (from Browser info or Item Properties [cmd+9]) then you should be good.
I can see the artifacts in the "after" shot. I can't explain it but you may get a better result if you complete the edit, export as a self-contained clip and bring it back in for titling.
This isn't much... hope it works.
OK, on point one: it's not that the Canvas is on a different screen, it's that the image is being displayed way beyond ideal resolution and will (at a minimum) look mushy. I don't think it's not the cause of your problem but it does throw additional sh*t into the game...
Point two: All interlaced video has two fields (progressive video only had one) but the default dominance is different between SD and HD formats. SD uses lower field dominance while interlaced HD (like 1080i) usually uses upper field dominance. I say usually because it seems like every camera maker has their own version of HDV and some of the other formats. I'm not sure what happens to the field dominance when you downconvert HD to SD in camera, so that's the question mark.
Thirdly, Properties (Cmd +9) is just a keyboard shortcut to look at all the specifics of an item, in this case the video specs. I'm not saying that a recapture is necessary, just that you need to verify the clip specs to make sure they are the same.
Finally, I want to add that in order to really assess what you have, you need to view it in its intended manner. That is, you should be viewing the footage on a PAL production monitor. Viewing interlaced video on a progressive screen (pretty much everything out there except CRTs) will reveal some undesirable artifacts (and the opposite is true as well.) What bothers me here is that the video is interlaced raster images and the titling is progressive vector (pretty much the opposite) so something is going to have to give in the render. Your perception of what "gives" will be based on how you view it. That's why I suggested that you may want to pre-comp (export a completed edit and bring it back in for titling) the interlaced timeline. It is also possible to rasterize your titles (in Photoshop or a similar app) and see if that helps.