LabVIEW
Parse formula in Labview
As I recall, the native LabVIEW formula parser has a problem. Each evaluation, all input variables are evaluated. This means that the more variables you have, the slower it gets. This is really something that you want to be done up front. So if the native parser is slower then yours, simply add more variables until it's slower then yours 
.
Another problem with the native parser (IIRC) is that only the first three characters are used to index the variables. Probably a 20 year old design decision, but really not good.
Search LabVIEW like a graph! 
I did a long time ago a completely reworked version of the formula parser here.
It's basically a complete rewrite from scratch up and it was able to perform pretty quick back in LabVIEW 6.0. I used this as a base in several applications since, where I did improve on the original design, but that is to intertwined with the actual application logic that I could post it now. As it was created in LabVIEW 5.1 days it does not use any classes and some of the architecture is by nowadays standards a bit arcane.
One of the features is that it has a parser part which translates the actual formula into a stack based UPN notation tree, and the actual expression evalution part which calculates the formula result. It supports arbitrary variable names with the restriction that a variable name needs to start with an alphabetic character and then can contain any alphanumeric character and the underscore as many times as you wish.
It still is a useful starting point for your own work, although various optimization work in the LabVIEW compiler itself has made the original huge performance difference between the NI formula parser and mine a lot smaller.
I haven't looked at the current NI implementation but running a similar test to yours on my machine with my libraray (LV 8.0, Windows 10, Core i7 2GHz) seems to indicate that they must have considerably reworked something in there as the timing on my library seem pretty similar to the one you quote for the NI library.
1000000 * evaluation only = 1.25 us per iteration
10000 * (parsing + evalution) = 104 us per iteration
There is some jitter when running it multiple times but the values stay fairly consistent.
Included is the 8.0 compiled version of the VIs..
Hmm your speeds look better than mine, despite having not that different hardware.
Mine is a "Core i7-6600 CPU @ 2.60 GHz 2.81 GHz" according to the System Control panel, but it is a Laptop currently operating on battery, so I suppose that could make a difference as the CPU is very likely not operating at max speed in that case.
