Model validation: Pearson residuals vs fitted values - when should I switch model due to heteroscedasticity?

Question

I'm carrying out GLMMs at the moment on coral reef fish count data against a number of covariates. My response variable data is pretty highly zero-inflated (66.7 percent). I experienced overdispersion and so after checking some other things, moved up from a Poisson to Negative binomial. I simulated my data (created 1000 simulated datasets) to see if the model could handle 66.7% zeros and found that value to be totally fine (see figure 2). However, when I did my model validation I found that my pearson residuals vs fitted graph (figure 1) shows some heteroscedasticity (slightly cone-shaped). I know that in reality these graphs aren't usually perfect but is mine problematic? And would switching to a Zero-inflated negative binomial fix the issue especially since I don't have an issue with zero-inflation.

Here are the graphs.... Thanks!