[ebbv]

I always assumed because the point of the diagonal arrow cursor is located at 0,0 in the image, making the origin location of the mouse cursor image and the click point of the arrow the same. Whereas with any vertical arrow cursor, the click point would no longer line up with 0,0.

[eterm]

In the windows 3.1 tool I mentioned above, you could actually choose a custom click point to be any point on the cursor.

One good prank was to set someone's click point to outside the cursor.

[StavrosK]

Oh yeah, I remember that. ImgEdit or something? It was great, I spent so much time making the hand/finger cursor into a middle finger rather than an index.

[kalleboo]

That could have been an issue with the PARC machine that OP claims started the trend, but with any OS after that (including the original Mac), you can specify the cursor hotspot when designing the cursor (look at your I-beam cursor for instance)

[blueskin_]

This is true for most default cursors, but custom cursor file formats let you change the coordinates of the active pixel (see: 'left-handed' mirrored arrow cursors, or crosshair-type ones.). The link 'finger' cursor on most OSes is a default example where the active pixel isn't the top-left one.

[Zecc]

For resize arrows it's pretty much a must.

[Sharlin]

The caret cursor over textfields as well.