Maybe do manual version of Breush Pagan test for heteroscedasticity:
1. Save the SQUARED residuals from estimated model with k independent variables.
2. Then regress squared residuals on the k independent variables and a constant.
3 Use N*R-squared from this regression as test statistic with distribution chi(k) ... (H0: homoscedasticity)
I have no idea whether this is statistically valid for an ordered probit model. However it is the bptest done manually....However the Breusch-Pagan test is a Lagrange Multiplier test and Verbeek writes in "A guide to modern econometrics" that a test for normality can be done according to a standard LM-test framework so I suppose the same is true when testing for heteroscedasticity....