I don't think it's caused by the inlet manifold directly - the alloy plenum has a similar dip, and it varies car to car.
I suspect it's a function of cam profile and timing.
Campbell's vs mine
Here you can see my car in orange - it's got the same cams but has alloy plenum and much bigger valves and more gas flow. As you can see it's actually worse than Campbell's across most of the rev range (not sure why, possibly down to standard exhaust back box on my car, more likely because of stupid alloy plenum, poor port matching and perhaps incorrect cam timing (no verniers)).
Anyway, point is that it too has a dip, though a little higher up the rev range (I wouldn't read too much into the exact position and size of the dip - remember the chart is only plotted and captured at 500RPM intervals, so both cars could have a very similar looking curve in reality).
One advantage of the hub-attached torque sensing "rolling road" that I visited with Hans Baumhardt (LOT racing driver) over xmas is that you get very closely spaced samples. We could probably persuade DASTEK to do a run with higher resolution capture, but they're very careful not to give you the raw data - not sure why they care
What's really happening in the engine, I think, is that 0-3,000 RPM the torque increases with engine speed as the inlet duration becomes better matched to the piston speed - there is less and less air/fuel being spat out on the compression stroke while the inlet valve is still open, so cylinder fills better.
3,000-3,500 torque is dropping because the effect of the inlet duration has peaked; the cylinder can no longer fill in the time available so torque drops off.
3,500-5,500 RPM the torque increases/plateaus as the overlap on the cams comes into play (the exhaust scavenging is helping to charge the cylinder). If you had no overlap at all, torque would continue to fall.
Changing the cam timing to increase overlap (advance the inlet a little) might make scavenging come into play a bit sooner, but would cost torque lower down I think (you would push exhaust gas into the plenum which would then be recycled thus reducing torque). However, each car is unique and there is a good degree of variance in the cam timing, so a vernier to fine tune it would make sense - basically advance it until torque drops off, then retard a tad. Similarly the exhaust cam could be tuned - retard it until torque drops off, then advance it a tad.
The best way of nulling out the dip is probably to fit a better exhaust manifold that would increase the scavenging effect - this would probably give a few extra BHP somewhere in the rev range and help pull up torque below 3,500RPM. It would have no effect on performance below 3,000RPM I think as the cylinder can empty and fill completely in the available cycle time.
There's a lot to be said for continuously variable cam timing!
Cheers,
Robin
) 
