I think its all about the jaggies. That is to say the stair-step effect you see on diagonal lines.
Basically the computer will output at whatever resolution you ask it to so if you are connected via VGA or DVI A/D in computer mode on a display with that level of resolution then you are being upscaled, but I'll caveat this in a second.
If on the other hand you are running a PC display at a higher resolution than your display is capable of then no matter your desktop resolution you are probably downscaling (so this is most true if you are using composite/svideo/component). The screen will do its bes, but it doesn't have the ability.
Now if you are upscaling the computer can do either a crude job or a good job. A good job needs processing power or clever video circuitry.
Lets make some assumptions to make things easier.
Imagine a still frame of video is a jpg picture at say 640 x 360. And lets also assume that your desktop resolution is 1280x720. Now when you have simplistic upscaling the computer simply draws each pixel twice horizontally and vertically. This does not change the effective resolution and you end up with a picture that has more pronounced stair stepping.
Now if you have smart upscaling then what happens is that the computer tries to "interpolate" the pixels that is rather than simply mapping a single pixel to a square of 4 pixels the same colour, it blends the colours with the surrounding pixels and ends up with a newly shaded block the following rather crude diagram may help a little
Original pixels
OOOXXO
OOXXOO
OXXOOO
Crude rescale
OOOOOOXXXXOO
OOOOOOXXXXOO
OOOOXXXXOOOO
OOOOXXXXOOOO
OOXXXXOOOOOO
OOXXXXOOOOOO
Good rescale
OOOOOOXXXXYO
OOOOOYXXXXOO
OOOOXXXXYOOO
OOOYXXXXOOOO
OOXXXXYOOOOO
OYXXXXOOOOOO
I've highlighted the Y pixels to show what difference the clever scaling makes. Notice the smoother diagonal line.
This effect is also called antialiasing.
Now smart upscaling also I believe tries to sharpen the picture as well. This means it needs to both smooth the picture and try and emphasise detail at the same time. This all takes CPU time.
So MCE on its own won't do a good job. dscaler, ffdshow and the rest are all there to do the job in a smart way. The same with purevideo.
Now the example I gave above is a simple doubling because it is the easiest to illustrate. Obviously the scaling is much more fractional in reality and necessity for a good method is essential.
The reason it is processor intensive is that for example antialiasing is achieved by supersampling. Effectively it scales the picture by 2, 4, 8, 16 and then re-averages the picture down to the required resolution. This averaging creates the second part of the definition of superscaling eg 16bit, 24bit, 32bit. This same technique is used in games and is what people talk about with x2AA, x4AA
Now remember that a DVD stream is 720x480 or 720x576 with fields (half frames). So it is more complicated still,as piecing together the right frames to construct that original picture (movie frame) has to happen first. It then gets scaled to any number of resolutions, and don't forget that widescreen means the whole picture is a bit distorted, and that's before you get to displays with non-square pixels.
So the idea for a quality set up is to know your displays true resolution, send the signal to the display at that resolution and perform all the correct scaling upfront.
So if you want good upscaling you must install it with software on MCE or the media player of your choice by using a filter/codec such as purevideo, ffdshow, dscalers.
Hope that helps!
Cheers
Rajiv