I landed the patch of "Reuse unmatched service worker processes for navigation".
https://chromium-review.googlesource.com/c/535220/
So now, the PlzNavigate's navigation will reuse the 'unmathced' service worker process.

But while I'm inspecting the logic of PlzNavigate, I found a edge case where two processes are created for service worker and navigation.

When PlzNavigate is enabled, a SiteInstance and a RenderProcessHost is created when NavigatorImpl::NavigateToPendingEntry() is called.

[1] NavigatorImpl::NavigateToPendingEntry()
 => NavigatorImpl::NavigateToEntry()
  => NavigatorImpl::RequestNavigation()
   => FrameTreeNode::CreatedNavigationRequest()
    => RenderFrameHostManager::DidCreateNavigationRequest()
     => RenderFrameHostManager::GetFrameHostForNavigation()
      => RenderFrameHostManager::CreateSpeculativeRenderFrameHost()
       => SiteInstanceImpl::GetProcess()

But RenderProcessHostImpl::AddExpectedNavigationToSite() is called when NavigatorImpl::OnBeginNavigation() is called asynchronously after NavigateToPendingEntry() is called.

[2] NavigatorImpl::OnBeginNavigation()
 => NavigationRequest::BeginNavigation()
  => NavigationHandleImpl::WillStartRequest()
   => NavigationRequest::OnStartChecksComplete()
    => RenderProcessHostImpl::AddExpectedNavigationToSite()


If the service worker is speculatively started from Omnibox, ServiceWorkerProcessManager::AllocateWorkerProcess() may be called between [1] NavigatorImpl::NavigateToPendingEntry() and [2] NavigatorImpl::OnBeginNavigation().
In that case, AllocateWorkerProcess() creates a new process.

I think it is a rare case. But I want to file a bug not to forget about it.