With --isolate-extensions, forking (via OpenURL) for extensions is not needed.

chrome/renderer/chrome_content_renderer_client.cc

Index: chrome/renderer/chrome_content_renderer_client.cc
diff --git a/chrome/renderer/chrome_content_renderer_client.cc b/chrome/renderer/chrome_content_renderer_client.cc
index 041c18ce689cb458166d7c5aa143a60a5ad89514..3282730c9419f063c8d3947b1f2116db0ddcac59 100644
--- a/chrome/renderer/chrome_content_renderer_client.cc
+++ b/chrome/renderer/chrome_content_renderer_client.cc
@@ -1139,16 +1139,6 @@ bool ChromeContentRendererClient::ShouldFork(WebLocalFrame* frame,
     return true;
   }
 
-  // TODO(lukasza): https://crbug.com/650694: For now, we skip the rest for POST
-  // submissions.  This is because 1) in M54 there are some remaining issues
-  // with POST in OpenURL path (e.g. https://crbug.com/648648) and 2) OpenURL
-  // path regresses (blocks) navigations that result in downloads
-  // (https://crbug.com/646261).  In the long-term we should avoid forking for
-  // extensions (not hosted apps though) altogether and rely on transfers logic
-  // instead.
-  if (http_method != "GET")
-    return false;
-
   // If |url| matches one of the prerendered URLs, stop this navigation and try
   // to swap in the prerendered page on the browser process. If the prerendered
   // page no longer exists by the time the OpenURL IPC is handled, a normal