There's no cosine "error" when the ball is moving downward or sideways due to a combination of gravity/spin. The gun is reading the velocity going toward the gun. The overall speed of the ball is the vector sum of the forward and sideways/downward movements but the gun in effect only reads the velocity toward the gun.
For example, a fastball released dead level at 90 mph which reaches the plate going 82 mph will be traveling about 120.27 fps forward and 8.56 fps down as is crosses the plate. It's going down at about 4 degrees as it crosses the plate. The total speed is a whopping 120.57 fps but the gun will read the 120.27 fps that it is moving forward. It is going toward the gun at 82 mph so that's what the gun reads. (Note: This is different from what Midlo was saying. He was saying that some of the forward velocity was being drained off to create velocity in the sideways/downwards direction. This difference is real but tends to be fairly negligible as the velocity change due to spin alone isn't that much. Also in general, the more spin there is the less drag there is so they tend to offset.)
A curveball might be going downwards at a much larger angle as it crosses the plate and the resultant speed of the ball may be a bit higher than the gun reading, but the gun is still reading the velocity toward the gun correctly.
Cosine error only comes into play when the gun is well offline from the path of the ball and it does have to be pretty far offline to be significant.
I often gun the kids from a spot that is about 10' above the plate and only about 20' behind the plate. That means a cosine error of about .77 mph on an 85 mph fastball and that I'll get a 1 mph low reading off the gun more often than not if the ball is released dead level.
Now let's look at the reading I'd get if I was that close to the plate as the ball crosses the plate. Now with a height of 10 ft and a horizontal distance of 20 ft I'd get an angle of 26.6 degrees and a ball with a forward velocity of 80 mph as it crossed the plate would give me a reading of about 72 mph due to cosine error. However, the movement of the ball wouldn't be contributing to that.
So there is a possibility of cosine error affecting gun readings at the plate if the gun is not very far behind the plate at all and offline a significant amount but the movement of the ball doesn't contribute. The reality is that it is very rare for there to be a chance to gun from so close and so far offline that it significantly affects the gun readings at the plate.
One of the funniest examples of cosine error I ever saw was at an A's vs. Giants game where they had the gun setup at almost a 45 degree angle (i.e. 30 feet in the direction of the ball and 30 feet to the side) from the path of the ball. People are almost always surprised at how slowly they throw but the readings they were getting were ridiculously low. That was a real ego buster.