I think it can be more subtle than just the obvious 'Essential Maths'. I've done a fair amount of graphics coding - 2D and 3D, mainly non-gaming, quite a lot heavily fractal-based - and got by fine on the maths I'd been taught at college plus self-education of anything else I needed (like most coders I guess I'm pretty good with maths) - so the usual Linear Algebra, Logic, Basic Calculus etc - and I've never felt particularly constrained by what I knew or could teach myself.
However at one point I had the opportunity to work with a young guy who'd just done a joint maths/computing degree to develop an atmosphere engine (one that models light decay and other atmospheric/light interactions). What particularly impressed me was his familiarity with a vastly wider range of mathematical functions than a 'normal' coder - and consequently when a particular behaviour was required he was able to reach for a just the right mathematical function and implement that to solve a problem in hand. None of the maths he used was actually that complex - I could understand it fine when shown it - but rather because he knew his mathematical vocabulary with such depth and breadth he was able to reach for 'le mot juste' effortlessly whenever he needed it rather than floundering around looking for it like most of us.
So I have a suspicion that, certainly in some problem domains, and possibly a wider selection than is obvious, a greater familiarity with maths than is common knowledge would lead to the implementation of better algorithms. We may only see a problem in terms of the tools in the toolbox and don't actually realise we're missing anything we don't have.