Title: Is it forbidden for multiple Framebuffers to reference the same Renderbuffer in OpenGL?

I am rendering a large number of points (~10 million) and drawing a rectangle in front of them to occlude some of the points.

image

As more points get occluded, performance should improve. However, in certain situations, I observed that performance actually drops as more points are occluded, compared to when more points are visible. I have confirmed that the occluded points do not execute pixel shader operations.

Here is the relevant code:

glBindFramebuffer(GL_FRAMEBUFFER, FBO);
glClear(GL_DEPTH_BUFFER_BIT);

GLuint newFBO;
glGenFramebuffers(1, &newFBO);
glBindFramebuffer(GL_FRAMEBUFFER, newFBO);
glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, colorRenderBuffer);

constexpr bool bShareDepthBuffer = true;
if (bShareDepthBuffer)
    glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_RENDERBUFFER, depthRenderBuffer);
else
{
    GLuint newDepthRenderBuffer;
    glGenRenderbuffers(1, &newDepthRenderBuffer);
    glBindRenderbuffer(GL_RENDERBUFFER, newDepthRenderBuffer);
    glRenderbufferStorage(GL_RENDERBUFFER, GL_DEPTH24_STENCIL8, m_width, m_height);
    glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_RENDERBUFFER, newDepthRenderBuffer);
}

glBindFramebuffer(GL_FRAMEBUFFER, FBO);
DrawBaseObjects2(*camera);

FBO is the primary framebuffer that I normally use, and colorRenderBuffer and depthRenderBuffer are attached to it.

During rendering, I create a new framebuffer, and only when bShareDepthBuffer = true, I attach the same depth buffer (depthRenderBuffer) to it.

When bShareDepthBuffer = true, the performance drops as more points get occluded, even if the newly created framebuffer is not used afterward. However, if I call glClear(GL_DEPTH_BUFFER_BIT); after glFramebufferRenderbuffer, the performance issue disappears.

If I instead create a separate depth buffer for the new framebuffer (bShareDepthBuffer = false), performance improves as expected, with better stability when more points are occluded.

Why does sharing the same depth renderbuffer across multiple framebuffers lead to a performance drop? The post states that binding the same renderbuffer to multiple FBOs should work. Is there something I might have overlooked?

Thanks in advance!