Ken's answer is close but not quite right.
On the real arcade hardware the signal sent to the CRT monitor is read directly from RAM as the electron beam scans over the screen. That means that whatever is shown on screen is whatever is in RAM at the moment the beam passes that area. The game's CPU takes a few frames to draw the entire play area (the girders, ladders and other graphics) and during those frames the video signal will be created from the partially drawn data in RAM.
It's actually even more subtle than that though. Look carefully at this frame:
You can see that girders 3, 4 and 5 on the right are partially drawn. Ignore the ones on the left, they are actually from the next frame. The beam is scanning left to right, and the CPU is rendering the girders top to bottom. The strange diagonal tear is actually due to the way the camera scans top to bottom.
Girders 1 and 2 are fully rendered by the time the beam scans over them. Girder 3 was about half way drawn, so the left half of it is not visible. At the time the beam was scanning over that area and the video signal was being generated the graphics for that girder had not been written in to RAM yet.
Girder 4 was drawn next, and you can see that the beam has moved a little further to the right by the time it began rendering. Same thing with girder 5.
MAME works a little differently. Because it is both difficult to implement and requires a great deal of processing power, rather than simulate the way the video signal is generated as the beam moves the emulator generates the entire image simultaneously. When the emulated beam is at the left hand side and about to begin generating the display, MAME reads the entire video RAM in one go. So the effect of the beam reading the RAM as it moves across the screen is lost.
Some emulators do simulate this effect, notably Amiga emulators.