A memory leak is when memory is allocated and retained, so that the garbage collector cannot reclaim it.
But your screenshot shows that the GC works efficiently, as memory can go down to around 30 MB.
So it doesn't look like a memory leak.
Now, you have to know that when running a Java program, you can specify the starting memory, which is always allocated, and the maximum memory to use. As you can see with the orange area growth, the memory manager can allocate more memory (up to the specified maximum) if the memory grows a lot before being collected.
AFAIK, the memory allocation never goes down.
You specified a maximum memory usage of nearly 1 GB, and the current heap is below 400 MB, still far below the maximum. You have to add the perm gen space to get the total consumed memory.

I run it again with the command line option -Xmx512m, and the results are still weird (at least for me):

Are you using any native libraries?  The one time I've known memory leaks not seen in VisualVM is with native buffers not being deallocated.

Are you getting as far as seeing an OutOfMemoryError?

In addition to the libs of the first post, I'm using P3D renderer, so I suppose I'm using jogl libraries, which I think have native parts?

" I'm using P3D renderer, so I suppose I'm using jogl libraries, which I think have native parts?"
In Processing 1.5, the P3D renderer doesn't use OpenGL. Are you using Processing 2.0?
And yes, Java OpenGL has (lot of) native parts.

2.0, I tested a few versions, a5, a6 and revision  9765 , all with similar problems.

I run a test of the sketch, with enough test time to see it achieved a little more than 2 GiB (according task manager), and then it reduced the footprint to 1.5 GiB (first significant memory reduction since sketch started).

If you never see an OutOfMemoryError and you eventually see the memory footprint reduce, stop worrying about it unless you have a very specific problem.  If you Google it you'll find lots of questions and articles about it, which generally boil down to the Task Manager being a really bad way of measuring actual memory usage.  Most seems to be down to the way the OS manages memory, pre-caches things, etc.  It's the OS doing its job, and some of that extra memory should get reallocated when needed.

I've found that I'm getting a memory leak that is somehow originating from Control P5.  When I comment out my controlP5.draw() function my memory stops creeping (eventually leading to an out of memory error). I have a huge controlP5 function, which I expect has something to do with it.

some code example and some specs (which processing version, renderer, os) would be helpful, if too large you can email them to be or create a pastebin.com document?

I had the same issue today.       working in Processing 1.5.1

solved it by downloading the latest version of the Library control P5