When running layout tests, we leak a WKWebViewConfiguration every times we reconfigure between tests. I believe I've tracked this leak down to line 477 in WKWebView.mm. Note that the biggest problem with leaking a WKWebViewConfiguration is that a WKWebViewConfiguration holds a strong reference to the ProcessPool. In effect, this means that leaking a WKWebViewConfiguration will create orphaned processes.
<rdar://problem/35230365>
I mis-diagnosed this leak. My reduced test case did a bad job at catching the autorelease of WKWebView, which cleans up the WKWebViewConfiguration I thought was being leaked.