I found the problem:

You have put <customErrors mode="Off" /> inside </system.web>

I can not explain exactly why, maybe one treatment will catch the other, but in my case it worked.

Yes. Some.

The correct configuration would initially look like this:

<configuration>
...
  <system.web>
  ...
    <customErrors mode="On">
      <error statusCode="404" redirect="~/Views/Erro/404.html" />
    </customErrors> 
  ...
  </system.web>
...
</configuration>

But I particularly do not like it because it goes against the MVC architecture. I'm more the Actions approach of Controllers to produce more interesting error screens:

<configuration>
...
  <system.web>
  ...
    <customErrors mode="On" defaultRedirect="~/Error/Index">
      <error statusCode="404" redirect="~/Error/NotFound"/>
      <error statusCode="403" redirect="~/Error/BadRequest"/>
    </customErrors>
  ...
  </system.web>
...
</configuration>

The problem is that this mode causes the transfer to return a code 302, and soon after a code 200, which hurts the HTTP protocol, since we had an error (ie codes 400 onwards), then we need adapt again.

<configuration>
...
  <system.web>
  ...
    <customErrors mode="On" redirectMode="ResponseRewrite" defaultRedirect="~/Error/Index">
      <error statusCode="404" redirect="~/Error/NotFound"/>
      <error statusCode="403" redirect="~/Error/BadRequest"/>
    </customErrors>
  ...
  </system.web>
...
</configuration>

Another problem: this does not work right because ASP.NET MVC uses Server.Transfer to redirect to static content, not dynamic. In the end, the configuration would look like this:

<configuration>
...
  <system.web>
  ...
    <customErrors mode="On" redirectMode="ResponseRewrite" defaultRedirect="~/Views/Erro/404.html">
      <error statusCode="404" redirect="~/Error/NotFound"/>
      <error statusCode="403" redirect="~/Error/BadRequest"/>
    </customErrors>
  ...
  </system.web>
...
</configuration>

Anyway, I do not really like either way. I'd prefer one, proposed by this answer here :

Global.asax.cs

public class MvcApplication : HttpApplication
{
    protected void Application_EndRequest()
    {
        if (Context.Response.StatusCode == 404)
        {
            Response.Clear();

            var rd = new RouteData();

            rd.Values["controller"] = "Errors";
            rd.Values["action"] = "NotFound";

            IController c = new ErrorsController();
            c.Execute(new RequestContext(new HttpContextWrapper(Context), rd));
        }
    }
}

ErrorsController.cs

public sealed class ErrorsController : Controller
{
    public ActionResult NotFound()
    {
        ActionResult result;

        object model = Request.Url.PathAndQuery;

        if (!Request.IsAjaxRequest())
            result = View(model);
        else
            result = PartialView("_NotFound", model);

        return result;
    }
}

I'd say that's the most decent way to handle HTTP errors in an MVC application.

Incidentally, it pays to read the answer I put here. It explains more fully all the problems of using HTTP errors by configuration.